Only seeing speedup with SGGLSE

Support for issues specific to the Linux operating systems.

Only seeing speedup with SGGLSE

Postby pcatalano » Thu May 05, 2011 8:45 am

Hello,

I recently downloaded and installed the free version of CULA to evaluate it. I ran the benchmark but found that it in most cases it performed worse than than MKL. These the are the specs of the machine I tested it on:

Processor: AMD Opteron 6174, 48 cores
Memory: ~125 GB
Graphics: GeForce GTS 250
Nvidia Driver: 270
Cula: R11

The graphics card is on the low end as far as CUDA goes, but I figured I'd see no change in performance using it, instead of a slow down. Is this system configuration just one that wouldn't benefit from using the GPU? Is there perhaps anything else I could do to get CULA to run smoother? Does the premium version provide better benchmark results, or does it just contain more functions?

Thank you for your time!

(Benchmark results below)

Code: Select all
Initializing CULA...
Initializing MKL...

Benchmarking the following functions:
-------------------------------------
             SGEQRF
             SGETRF
             SGELS
             SGGLSE
             SGESV
             SGESVD
-------------------------------------


     -- SGEQRF Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096       3.27       1.54    0.4712
5120       5.99       2.35    0.3929
6144       9.03       3.46    0.3830
7168      15.33       5.13    0.3346
8192      22.48       7.21    0.3208

     -- SGETRF Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096       0.67       0.60    0.9046
5120       1.18       0.91    0.7752
6144       1.48       1.44    0.9743
7168       2.88       1.94    0.6757
8192       4.16       4.23    1.0170

     -- SGELS Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096       3.33       1.84    0.5527
5120       5.99       3.12    0.5205
6144       9.96       5.05    0.5074
7168      14.96       6.84    0.4572
8192      21.83       9.10    0.4168

     -- SGGLSE Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096       3.71       5.40    1.4558
5120       6.79       7.68    1.1302
6144      10.09      11.19    1.1094
7168      16.58      25.05    1.5114
8192      24.45      45.57    1.8638

     -- SGESV Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096       0.85       1.00    1.1788
5120       1.44       1.36    0.9418
6144       1.86       2.16    1.1653
7168       5.95       3.22    0.5412
8192       4.78       7.55    1.5796

     -- SGESVD Benchmark  --

Size   CULA (s)    MKL (s)   Speedup
------ ---------- ---------- ---------
4096      78.76      76.72    0.9740
5120     157.35     137.45    0.8735
6144     248.89     219.24    0.8809
7168     399.52     378.26    0.9468
8192     599.49     578.02    0.9642
pcatalano
 
Posts: 1
Joined: Tue Apr 19, 2011 7:34 am

Re: Only seeing speedup with SGGLSE

Postby kyle » Thu May 05, 2011 11:53 am

You are comparing four brand new, top of the line, 12-core processors to a single entry level CUDA card; your results are pretty much as expected.

On the bright side, this demonstrates the high FLOPS/$ available from the GPU as a $100 GPU can keep up with a $10k (I'd estimate) system :)

For a more appropriate display of performance, check our out graphs located at our performance page: http://www.culatools.com/features/performance/
kyle
Administrator
 
Posts: 301
Joined: Fri Jun 12, 2009 7:47 pm


Return to Linux Support

Who is online

Users browsing this forum: No registered users and 1 guest

cron